1. Welcome to:
Real-Time Optimization:
Putting Facebook User Attributes to Work
- Going Beyond A/B Testing and User Segmentation
- Optimizing Open Graph
Alan Avidan − Executive Director
alan@BeesAndPollen.com
@beesandpollen
2. We’ll Cover:
1. The Playground: Games/Apps/Campaigns
2. Which User Attributes Can You Use for Optimization?
3. Predictive Best-Fit Optimization, and How Does it Lift
KPIs like Revenue, Virality, Engagement, Retention
4. Traditional Optimization Tools:
Analytics, A/B Testing, User Segmentation
5. Open Graph Optimization with Predictive Best-Fit
3. • Lots of Successful Apps, Games and Campaigns with
Millions of (Individual) Users
• Low Retention, Low %Pay, High User Acquisition Costs
• Notifications/Posts Can Become Spammy and Blocked
• KPIs Under Pressure – Need Lift - Perform or Perish!
• Vast Amounts of User Attributes
4. Terminology
• Attributes
• Elements (Events/Decision Points)
• Options (Variants)
Low Range High Range
5. User DNA - Attributes Sources
Facebook attributes: Friends, Influence, Likes, Interests, Posts, Events, etc.
Open Graph: scores, achievements, published stories, custom actions, etc.
Behavioral attributes: level, spending, score, health, custom, etc.
Session attributes: time of day, day, duration, etc.
Geo-Demographic attributes: age, gender, education, country, etc.
3rd Party attributes: income level, education, etc.
6. Predictive Best-Fit – Core Concepts
Predictive Best-Fit Algorithms Find Correlations
Between User DNA and Conversions
User Social, open-graph Predictive Best-Fit
DNA Generation Algorithm Real-
User and Behavioral Data
Time action
8. A/B Testing
Define options Split traffic Measure results Deploy winner Max Result
high range
Low range
High range
9. A/B Testing – Bottom Line
Upside
• Conceptually simple and understandable
Can achieve good results – up to a point
Downside:
• One-size-fits-all
• Results may deteriorate over time
10. A Priori Segmentation
Define segments Define Options and rule Result
base
Low range
high range
11. A Priori Segmentation
Upside
• Can be effective if segmentation was meaningful
Downside
• Segments are predefined and cannot be changed
during the analysis
• Different elements might require different segments
• Hard to scale in terms of data-set and number of
elements
• Hard to fine-tune
12. Clustering Segmentation
Define A/B test Segment users Deploy winner
options options based on result
Low
range
High range
13. Clustering Segmentation
Upside:
• Highest Lift
• Discover correlations you never knew existed
Downside:
• Requires storage of terabytes of data
• Need really smart people to work on it
• Effort = Very High
14. Predictive Best-Fit
• Can optimize in-app and open graph performance
• Automated end-to-end solution
(Acquire data, analyze, predict, enact)
• Machine self-learning
• Real-time
• No user history required
• Numerous data sources
• In full compliance with facebook privacy rules
• Deep new insights
Effort/Resources
15. Elements For Predictive
Best-Fit Optimization
Open Graph Engagement Retention
• Publish Yes/No? • Offers • Email
• Timing • Products • Message Timing
• Art and Copy • Content • Incentives
• Call-to-Action • Communications • Gifts
• Story
Virality Look & Feel Monetization
• Share Messages • Colors • Payment Page:
• Invite Friends • Graphics Ranges, Incentives
• Layouts • Shop Order
16. Open Graph Big Impact
SongPop Hits Major Milestones Just Three Months After Launch
• 25 Million unique players to date
• Has consistently received a coveted 5 start rating
• 4 million people play every day, and growing
Ford created an app that publish a story each time a user
customized his dream Mustang and then battle others’ model.
Although their goal was 2 million engagement they had more than
5 millions and more than 17,000 referrals.
Since revamping Open Graph stories with custom art and content,
BINGO Blitz got 20% more likes and comments on news feed
stories and 500% more unique clicks to the game.
The food finding and sharing app has seen a 3X increase in number
of visits and activities shared by helping people share the dishes
they want, try and ate with friends on Facebook
17. Open Graph Optimizations
1 2
3 4
5
6
1 2 3 4
Publish by User – Yes/No Story Image Landing Page
5 6
Action Verb Object Time
18. Open Graph 1 Publish by User – Yes/No
Yes No
Publish only by the right users!
19. Open Graph 2 Story
Post with the right content to engage the viewer
• Publish achievements the player unlocked
• Publish scores the player achieved
• Publish custom activities:
Jeff E. finished Level 4 on MyGame!
• Publish extended custom activities:
Jeff E. won a game against Chris on MyGame!
20. Open Graph 3 Image
Option A Option A
Image of song, leading to Image taken from to
clip game
Option B Option B
Image of genre, leading Image of real-world
friends to songs/albums landscape
recently listened to by user
Publish using the most effective creative
21. Open Graph 4 Landing Page
Option A
Landing page with the song playing
Option B
Landing page with the latest songs of that genre listened by
friends’
Option C
Landing page of that album with a discount coupon
Publish with the best landing page to convert the viewer
22. Open Graph 5 Action Verb Object
listen
Option A
Justin listened to [SONG X] by [SINGER-NAME] on Spotify
Option B
Justin listened to Classic [GENRE Y] music on Spotify
Publish the most effective actions and objects
23. Open Graph 6 Timing
Publish at the right time to get maximal exposure
Friends newsfeeds
24. The Last Word
Consider optimization if you wish to
become successful or stay relevant
Consider Predictive Best-Fit Optimization
All the Gain without the Pain
25. Welcome to:
Real-Time Optimization:
Putting Facebook User Attributes to Work
- Going Beyond A/B Testing and User Segmentation
- Optimizing Open Graph
Alan Avidan − Executive Director, Business Development
alan@BeesAndPollen.com
@beesandpollen